The map certainly has its flaws because of the lack of data that is available. Only statewide data is available from exit polls. For the map, I didn't have any real way of systematically modifying the formula between counties within a state that wasn't purely subjective.
I'm sure exit polls did poll urban whites and rural whites separately right? Oh but there is that suburban business that makes the whole deal subjective.
There is also Census data that's available for regression purposes.
Variables I would probably use are
- McCain 2008 vote - M
- Percent White - W
- Median income - I
- College education - C
- Percent Married Couples - W
- Dummy variable for urban/suburban and not - D
- Unemployment in October 2008 - Google Public Data - U
log(W) = Constant + M + W + C + U + W + D
Once you have results for every county you can easily do a high confidence interval like (>= 99.9)to see if the number of votes works out.
Another method, though extremely crude and overgeneralizing, is seeing the correlation coefficient of %McCain and %White.
I can probably try to work on a simple linear regression result of %White and %McCain in California just to see tonight.
Whats problematic though is if there's an interaction term or if a certain variable has a higher threshold than others. We could try squaring or multiplying variables.
